In game theory, all organisms want to maximize their benefit. Game theory can used as a tool that makes programming learning machines easier than it has been. The idea is that the AI will get points for every correct answer. And when the AI gives the wrong answer, the system gets zero points. That thing allows the simple model, that is easy to program for computers.
When the AI plays chess against people, it can have a system where each chess button has points. That allows the AI to model the game. The soldier can be one or two points. And the queen can be worth 1000 points. When the number of those chess buttons decreases, the system can get new numbers for those buttons. The winner of the game is, who gets more points, and loses fewer buttons. The lost button can decrease the points as the value of those lost buttons is.
Image: Quanta magazine
In classic machine learning models the computer selects the answers. That brings benefit to it for storing in its memory. But the problem has been to describe the answer. That brings benefits to the system. And one of the ways to tell the system, what kind of solution is right, is to use points to describe the right answer. The right answer brings up one point. And wrong gives zero.
Things like attack drones can use this kind of point system for target selection. The main battle tank can be 100 points. The truck can be 20 points. The system can select targets using this model. And it can try to maximize its points.
The AI can also know, that a deck of cards must not contain two similar cards. That allows the AI to remove non-necessary data from the data pool. If there is the same data, stored in many places, the AI can remove the superimposed data packets.
In that model, the computer can find the right answer from the databases like Wikipedia pages. The system can find the right keywords using the data that is stored in Wikipedia.
The idea is to make a model that is easy to use for making an algorithm. That allows the computer to select. Does it benefit from the answer or not? Game theory is a tool, that can use for many purposes. AI is one of the newest tools and the system requires a model, that is easy to use when people try to teach that tool.
When operators teach the machine they ask questions, and the machine must find the answer. All AI's can be small-size specialized AI's. But the idea in the artificial general intelligence is that those AI's operate in one domain. That makes those systems effective and easy to expand.
https://www.quantamagazine.org/game-theory-can-make-ai-more-correct-and-efficient-20240509/
Comments
Post a Comment